Journal of Thi-Qar University No.3 Vol.

December/2008

Radial Basis ANN-Based Static Load Flow Analysis

in Iraqi National Super Grid (400 KV)
By
Dr. Baqer Turki Atiyah Al-lamey
Lecturer/ Al-Nasseriya Technical Institute
Abstract:This paper presents the application of Artificial Neural Networks for Loadflow analysis of Iraqi National super grid (400 KV network) using radial basis
neural network (RBNN) to handle the slow computational process of Error
Backpropagation Neural Network (EBPNN). The proposed method is fast and has
acceptable accuracy. Active and reactive powers of system buses excluding slack
bus, as well as the magnitudes and phase angle of slack bus are chosen as inputs to
the ANN. Phase angle and voltage magnitude of all buses excluding slack bus are
chosen as the outputs. Training data are obtained by performed a load flow
program using an iterative numerical method namely Decoupled load flow method
(DLF), same input and output parameters of RBNN are feed to the EBPNN. Load
flow analyses results achieve from two neural methods are compared with the
result of DLF method to illustrate the accuracy of result and the results are
compiled to form the training set. The proposed algorithm is applied and the
numerical results are presented in this paper in order to demonstrate the
effectiveness of this proposed algorithm in terms of accuracy and speed. It is
concluded that the trained ANN can be utilized for both off-line and on-line
simulation studies. The simulation programs implemented using 7.5-Matlab
programming language to obtain the satisfactory results.
( ANN)
( Radial basis NN) ( 400 KV)
. (Error Backprpagation NN)

EBPNN RBNN
.

. (Decoupled Load-Flow Method)
( EBPNN)( RBNN)( DLF)
311

Journal of Thi-Qar University No.3 Vol.4

December/2008

.
.MATLAB-7.5
Introduction:Load-Flow is very important and fundamental tool for analysis any power
system and is used in the operational as well as planning stage. Certain applications,
particularly in system automation and optimization of power system, require
repeated Load-flow solution. In these applications it is very important to solve the
load flow problem as efficiency as possible [1]. Almost all the known methods of
numerical analysis for solving a set of non-linear algebraic equations have been
applied in developing load flow algorithms. One or more desirable features to
compare the different load flow methods can be the speed of solution, memory
storage requirements, accuracy of solution and the reliability of convergence
depending on a given solution. Though, robustness or reliability of convergence of
the method is required for all types of applications, the speed of solution are more
important for on-line applications compared to the off-line studies [2].
Many researches on using Artificial Neural network in power systems had
been studied; Masanori, S. and et..al in [3] use artificial neural network to calculate
various load flow state in power system since the achievement results shown that
ANN network is a fast tool to perform the large scale power system calculations.
Nguyen, T. T. in [4] presents the development of a ANN networks architecture
which implements the Newton-Raphson algorithm for solving of power network
load flow equations. Jayendra Krishna and Laxmi Srivastava in [5] develop
counterpropagation neural network to solve Load-Flow problem under different
loading contingency condition for computing bus voltage magnitudes and angle of
the power system, the composition of the inputs variable of the proposed method has
been selected to emulate the solution process of a conventional load flow program.
Hung-Cheng Chen and et al in [6], proposed partial discharge classification using
neural networks and statistical parameters commercial partial discharge detector is
firstly used to measure the 3-D partial discharge patterns of epoxy resin power
transformers. Then, the gray intensity histograms extracted from the raw 3-D
partial discharge patterns are statistically analyzed for the neural-network-based
(NN-based) classification system. Noor Izzri Abdul Wahab and et..al in [7], presents
transient stability assessment of electrical power system using probabilistic neural
network (PNN) and principle component analysis. Transient stability of a power
system is first determined based on the generator relative rotor angles obtained
from time domain simulation outputs. In this paper we used Radial Basis and Error
Backpropagation neural networks with Decoupled load flow method to raise the
speed computation process and minimize the memory storage which required to the
Load-Flow analysis of Iraqi National Super Grid (400 KV network).
Static Load-Flow Analysis:The objective of Load-Flow analysis is to determine the voltage and its angle
at each bus, real and reactive power flow in each line and line losses in the power
system for specified bus or terminal conditions. The power flow studies are
conducted for the purpose of planning (viz. short, medium and long range
planning), operation and control. For the purpose of power flow studies, it is
311

Journal of Thi-Qar University No.3 Vol.4

December/2008

assumed that the three-phase power system is balanced and also mutual coupling
between elements is neglected. Variable associated with each bus of the power
system include four quantities viz. voltage magnitude Vi, its phase angle i , real
power Pi and reactive power Qi total 4n variable for n buses system. At every bus
two variables are specified, the remaining two can be found by solving the 2n power
flow equations. Out of these four quantities only two are generally specified at a few
buses and depending upon which two are specified, we have three categories of
buses, namely Swing Bus or Reference Bus, Generator Bus or PV Bus and Load Bus
th
or PQ Bus. From the nodal current equations, the total current entering the i bus
of m bus system is given by:-[5]
m

I i Y i1V 1 Y i 2V 2 ........ Y imV m Y ikV k

(1)

k 1

At i

th

bus, complex conjugate power will be

m

S P j Q V I V (Y V
*

or Pi j Q V i Y ik V k e
i

k 1

P Re al V Y V e
k 1

P V Y V
i

k 1

ik

Or

ik

j(

ik

ik

j(

ik

k 1

(2)

C os ( ik i k )

P V V [G Cos ( i k ) Bik Sin( i k )

i

k 1

ik

Similarly, the reactive power in the

m

k 1

bus will be

j(

ik

Qi V i Y ik V k C os ( ik
k 1

th

Im ajinary V i Y ik V k e
m

Or

(3)

V iV k [Gik Sin

k 1

i k )

( i k ) Bik Cos( i k )

(4)

Equations (3) and (4) are known as static Load-Flow equation, the load flow
equations used in Netown-Raphson method computation of voltage corrections are
given as:

V J 11
V J 21

J
J

P

Q
22

12

(5)

311

Journal of Thi-Qar University No.3 Vol.4

Where:

J 11

Pi

J 12

Pi

V k V k

J 21

December/2008

J 22

V k V k

The solution of Equation (5) provides the correction vector i.e. .s for all the
PV and PQ type buses and .Vs for all the PQ type buses, which are used to update
the earlier estimates of s and Vs. This iterative process is continued till the
mismatch vector i.e. .Ps for all the PV and PQ type buses and .Qs for all the PQ
buses become less than a pre-assigned tolerance value . As can be observed form
(5), during each iteration, Jacobian elements are to be calculated and its inverse is
also required, due to this fact, the Newton-Raphson method requires more time per
iteration.
Decoupled Load-Flow Method:In the strictest use of the Newton-Raphson procedure, the jacobian is
calculated and triangularized in each iteration. In practice, however, the jacobian is
often recalculated only every few iterations and this speed up the overall solution
process. The final solution is determined by the allowable power mismatches and
voltage tolerance at the buses. When solving large scale transmission power systems,
an alternative strategy for improving computational efficiency and reducing
computer storage requirements is the decoupled Load-Flow method, which makes
use an approximate version of the Newton-Raphson procedure. The principle
underlying the decoupled approach is based on two observations:- [8]

Change in voltage angle

at a bus primarily affects the flow of real power
(P) in the transmission lines and leaves the flow of reactive power (Q)
relatively unchanged.

Change in voltage magnitude V at a bus primarily affects the flow of

reactive power (Q) in the transmission lines and leaves the flow of real power
(P) relatively unchanged.

We have noted that the first observation states essentially that Pi

is much
j

larger than

, which we now consider to be approximately zero. The second

j

observation status that

is much larger than

j

Pi

, which is also
j

considered to be zero. Incorporation of these approximations into jacobian of

Newton-Raphson make the elements of the submatrices J 12 and J 21 zero, we are
then left with two separated systems of equations;

J
1

11

and Q

J V
1

22

(6)

To get fast computation in load flow solution, the decoupled load flow program has
been developed in this paper and run to generate several training / testing patterns.

311

Journal of Thi-Qar University No.3 Vol.4

December/2008

Neural Network Architecture:-[9]

A neural network functions like a human brain and is well known as a
universal approximator. Due to the capability of dealing with highly nonlinear and
complex problems, neural networks had been successfully applied to a wide variety
of real world problems. An artificial neural network has three essential segments;
an input layer, a hidden layer and an output layer. Depending on the complexity of
a solution process an ANN may require more than one hidden layers. Each layer is
made up of several neurons that transmit signal to the neurons at the next layer.
Each signal path has its own weight, known as synaptic weight. By selecting the
synaptic weights, an ANN can be made to simulate a complex solution process.
Activation functions are used at the hidden and output layers to transform the input
signals.
Radial Basis ANN: [10], [11]
The radial basis neural network consists of three layers, the input layer,
hidden layer and output layer. The nodes within each layer are fully interconnected
to the previous layer; the input variables are assigned to each node in the input
layer and are passed directly to the hidden layer without weights. The hidden nodes
contain the radial basis functions, and are analogous to the sigmoid function used in
the Error Backpropagation (EBP) networks. The radial basis function (RBF) is
defined by a center position and width parameter, the width of RBF unit controls
the rate of decrease of function.
The parameters of RBF unit are determine in three steps of the training
activity, first the unit centers are determined by some form of clustering algorithm,
the widths are determined by a nearest neighbour methods. Finally weights
connecting the RBF units and the output units are calculated using multiple
regression techniques.
Test system:The INSG-400KV (Iraqi National Super Grid) which is composed 22-Bus
and 29-Line has been used to test the proposed methodology. Bus BAJP make as
slack
bus
having
pre-specified
voltage
as
1.02 0 ,
BAJG,MMDH,HDTH,QDSG,MSUP and NSRP as voltage-controlled buses and the
remain buses as load buses. Figure (1) show the single line diagram of INSG
network, and the line data and power data of this network are tabulated in Tables
(1) and (2).[12]
Table (1): Line Data of Iraqi National Supper Grid System (400KV)
Line-Name

R
(P.U)

X
(P.U)

B
(P.U)

MMDH-MSL4(1+2)
MSL4-BAJp(1+2)
BAJP-BAJG
BAJG-KRK4
BAJP-BGW4
BAJP-BGW4
BAJP-HDTH
HDTH-QIM4
HDTH- BGW4
BGW4-BGN4

0.001436
0.004195
0.000022
0.001799
0.004832
0.004962
0.003446
0.002918
0.004845
0.000932

0.011768
0.034371
0.000197
0.016351
0.043931
0.045113
0.031323
0.02391
0.044049
0.008471

0.364392
1.064256
0.005837
0.484471
1.301651
1.336673
0.928083
0.740352
1.305153
0.250991

311

Journal of Thi-Qar University No.3 Vol.4

BGW4-BGS4
BGE4-BGN4
BGN4-QDSG(1+2)
BGE4-DAL4
BGE4-BGS4
BGS4-MUSP(1+2)
BGS4-KUT4
BGS4-KDS4
MUSP-BAB4(1+2)
BAB4-KDS4(1+2)
KDS4-NSRP
KUT4-NASP
KUT4-QRNA
NSRP-KAZG
KAZG-HRTP
HRTP-QRNA
DAL4-KRK4
BGS4-AMN4
AMN4-BGE4

0.001582
0.000288
0.000152
0.000867
0.001261
0.00122
0.000433
0.003075
0.000809
0.002326
0.003833
0.004321
0.006248
0.004393
0.001183
0.0013
0.001625
0.000823
0.000433

0.014383
0.00262
0.001379
0.00788
0.011465
0.010149
0.00394
0.027954
0.006734
0.019349
0.034849
0.039282
0.05713
0.039932
0.010756
0.01182
0.014775
0.007486
0.00394

December/2008
0.426101
0.077632
0.040859
0.23348
0.339713
0.318967
0.11674
0.82827
0.211651
0.608124
1.032565
1.163898
1.69273
1.18316
0.3187
0.35022
0.457775
0.221806
0.11674

Table (2): Bus-Bar's data of Iraqi National Supper Grid System (400KV)
Bus-Bar Name

V(p.u)

Pg(MW)

PL(MW)

QL(MVAR)

BAJP
BAJG
MMDH
HDTH
QDSG
MSUP
NSRP
HRTP
MSL4
KRK4
DAL4
BGE4
BGN4
BGW4
QIM4
BGS4
AMN4
KUT4
QRNA
KAZG
BAB4
KDS4

1.02
1.01
1.0
1.0
1.01
0.96
0.98
1.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0

0.0
500
500
500
242
600
650
380
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0

200
90
50
100
60
120
100
38
300
70
150
500
300
500
60
100
350
100
70
350
100
200

98
10
20
60
70
70
54
22
180
40
80
360
200
360
40
50
47
60
30
200
50
100

311

Journal of Thi-Qar University No.3 Vol.4

December/2008

Figure (1): Single line diagram of Iraqi National Super Grid system. [12]

Proposed Solution Method:Three layers Radial Basis and Error Backpropagation Neural Networks have
been used in this paper to provide the desired power system model. It is assumed
that the neural network architecture would be sufficient enough to recognize any
non-linearity in the simulated power system. The input vector of proposed neural
networks is 2N (N is the number of system buses) and the output vector is 2(N-1),
the size of hidden layer is based on educated trail and error approach; initially it is
identical on output layer size and then increased as learning/identification process
requires. Figure (2) shows the architecture of proposed neural networks, during the
training process of this network, a set of input values are fed to both neural
networks model and numerical decoupled load-flow iterative method, then the
output of three systems are compared to produce the error signals. The network
architecture specification is as the following: Input layer size equal to 44 neurons represent the slack bus phase angle and
voltage magnitude, and real and reactive powers of other buses.
Output layer size of equal to 42 neurons representing all buses (except slack
bus), estimated voltage magnitude and phase angle.
Hidden layer size equal to 56 neurons found by educated trial and error
procedure.

Input
slack

x
x

Hidden

Output

w
w

slack

11

i1

pv & pq buses

PV & Pq Busses

1p

w
w

im

remain buses

ih

mp

mo

remain buses

Basis Neuron
Figure (2): Architecture of proposed Radial Basis Neural Network
Simulation Results:After the implementation of conventional Load-Flow method program, the
training data of radial basis and error backpropagation neural networks are
311

Journal of Thi-Qar University No.3 Vol.4

December/2008

obtained then the conventional and neural programs are performed to achieve the
simulation results which are tabulated in Tables (3) and (4). Depending on the
results of Decoupled Load-Flow method which are considered as a main reference in
comparison due to its high accuracy with respect to neural methods, it was found
that the results of proposed neural networks method obtains a good average time
and accuracy that the neural network results not acceded in estimated values the
permitted error level percentage. Table (5) shows the results of solution methods in
the point of view of the execution time and accuracy. Figures (3) and (4) show the
training curve of proposed Error Back-propagation and Radial Basis Neural
Networks and voltage profile of test system calculated by three proposed methods
has been illustrated in Figure (5).
Table (3): Simulation results of Solution Methods from Voltage Magnitude Point of View
Bus Name

BAJG
MMDH
HDTH
QDSG
MSUP
NSRP
HRTP
MSL4
KRK4
DAL4
BGE4
BGN4
BGW4
QIM4
BGS4
AMN4
KUT4
QRNA
KAZG
BAB4
KDS4

Bus
No.
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22

DLF
Method

Bus

(P.U)

1.00000
1.0000
1.0000
1.0100
0.9600
0.9800
1.00082
0.9784
0.9882
0.9775
0.9827
0.9973
1.0041
0.9803
0.9885
1.0015
1.0023
0.9585
1.0188
1.0422
0.9667

EBPNN Method

Bus

(P.U)
1.0000
1.0031
1.0046
1.0140
0.9645
0.9833
1.0
0.971
0.977
0.974
0.990
0.993
1.000
0.983
0.984
1.001
1.00
0.964
1.020
1.040
0.967

RBNN Method

Error

%
0.0000
0.3100
0.4600
0.3960
0.4680
0.3367
0.0819
0.7560
1.1330
0.3580
0.7420
0.4310
0.4083
0.2754
0.4552
0.0499
0.2294
0.5738
0.1177
0.2110
0.0310

(P.U)
0.998
0.990
1.009
1.000
0.979
0.988
0.986
0.989
0.980
0.983
0.995
0.981
1.000
1.000
0.995
0.988
0.992
0.975
1.009
1.0392
0.972

Bus

Error
%
0.2000
1.0000
0.9000
0.9900
0.1979
0.8163
0.0819
1.0834
0.8297
0.5626
1.2516
1.6344
0.4038
1.7341
0.6630
1.3479
1.0276
1.7214
0.9619
0.8878
0.2379

Table (4): Simulation results of Solution Methods from Phase Angle Point of View
DLF Method
Bus Name

Bus
No.

BAJG
MMDH
HDTH
QDSG
MSUP
NSRP

2
3
4
5
6
7

(radian)

EBPNN Method

RBNN Method

Error

Error

0.0342
0.0963
0.0231
0.0765
0.0707
0.0734

(radian)
0.0346
0.0971
0.0226
0.0755
0.0694
0.0720

%
1.1695
0.8307
2.1645
1.3071
1.8387
1.9073

(radian)
0.0348
0.0973
0.0235
0.0733
0.0722
0.0752

%
1.7543
1.0384
1.7316
1.5686
2.1216
2.4523

311

Journal of Thi-Qar University No.3 Vol.4

HRTP
MSL4
KRK4
DAL4
BGE4
BGN4
BGW4
QIM4
BGS4
AMN4
KUT4
QRNA
KAZG
BAB4
KDS4

8
9
10
11
12
13
14
15
16
17
18
19
20
21
22

0.0671
0.0779
0.0542
0.0642
0.0601
0.0587
0.0589
0.0363
0.0888
0.0540
0.0417
0.0547
0.0429
0.0437
0.0218

0.0659
0.0766
0.0531
0.0629
0.0593
0.0593
0.0594
0.0365
0.0893
0.0538
0.0419
0.0538
0.0422
0.0444
0.0215

1.7883
1.6688
2.0295
2.0249
1.3311
1.0221
0.8488
0.5509
0.5630
0.3703
0.4796
1.6453
1.6317
1.6018
1.3761

December/2008
0.0689
0.0753
0.0554
0.0627
0.0611
0.0595
0.0596
0.0366
0.0895
0.0543
0.0422
0.0551
0.0433
0.0446
0.0222

2.6825
1.7971
2.2140
2.3364
1.6638
1.3628
1.1884
0.8264
0.7882
0.5555
1.1990
0.7312
0.9324
2.0594
1.8348

Table (5): The results of solution methods in the point of view of the execution time
and accuracy
Solution Method

DLF

EBPNN

RBNN

Execution Time(s)

4.46

1.48

0.65

99.618 %

99.117 %

98.659 %

98.436 %

Accuracy

Figure (3): Training curve of proposed EBP Neural Network

313

Journal of Thi-Qar University No.3 Vol.4

December/2008

Figure (4): Training curve of proposed Radial Basis Neural Network

Voltage profile of Iraqi National Super Grid

((400 Kv)

Voltage Magnitude

1.06
1.04
1.02
1
0.98
0.96
0.94
0.92
0.9
1

11

13

Bus No.

15

17

19

21

DLF Method
RBNN method
EBPNN Method

Figure (5): Voltage profile of test system calculated by proposed methods

Conclusions:Radial basis neural networks have been developed to solve power flow
problem in an efficient manner. In error backpropagation neural network the
training process is slow, and its ability to generalize a pattern-mapping task depends
on the learning rate and the number of neurons in the hidden layer. On the other
hand training of a radial basis neural network is very fast, at the same time the
generalization capability of the RBNN allows it to produce a correct output even
when it is given an input vector that is partially incomplete or partially incorrect.
The proposed solution method tested on Iraqi National Supper Grid (400 KV) using
7.5-MATLAB language in program implementation.

311

Journal of Thi-Qar University No.3 Vol.4

December/2008

References:1- Ray, D. Zimmerman and Hsiao, Dong Chiang, "Fast decoupled power Flow for
Unbalanced Radial Distribution systems", IEEE power engineering society for
presentation IEEE/PES winter meeting, New York, 1995.
2- Van Amerongen R., "A general purpose version of the fast decoupled Load
Flow", IEEE Trans. On power system, Vol. 4, No. 2, 1990.
3- Masanori, S. and et..al, "The calculation of electric power flow using neural
network systems", IEEE Japan, Vol. 114c, No.9,2002.
4- Nguyen, T.T., "Neural Network load flow for on-line static security
assessments", Energy system center, No. 10, 2002.
5- Jayendra Krishna and Laxmi Srivastava, "Counterpropagation Neural Network
for Solving Power Flow Problem", International Journal of Intelligent
Technology Volume 1 Number 1, 2006.
6- Hung-Cheng Chen and et..al, " Partial Discharge Classification Using Neural
Networksand Statistical Parameters", Proceedings of the 6th WSEAS
International Conference on Instrumentation, Measurement, Circuits & Systems,
Hangzhou, China, April 15-17, 2007.
7- Noor Izzri Abdul Wahab and et..al, " An Improved Method in Transient
Stability Assessment of a Power System Using Probabilistic Neural
Network", Journal of Applied Sciences Research, 3(11): 1267-1274, 2007.
8- Grainger, John. J and Stevenson, D. William .Jr, Power System analysis,
Neperdazan, Iran, 2003.
9- Rezaul Haque and N. Chowdhury, "An Artificial Neural Network Based
Transmission Losses Allocation for Bilateral Contracts", IEEE
CCECE/CCGEI, Saskatoon, May 2005.
10-Jain, T. Srivastava L. and S. N. Singh, " Fast voltage contingency screening
using radial basis function neural network", IEEE Transaction on power
systems, Vol. 18, No. 4, 2003.
11-Leonardo, paucar, " Artificial neural networks for solving power flow problem
in electric power systems", Electric power systems research, Vol. 62, 2004.
12- Al- sujada, Jwan S. Ra`afat Khadsim, Enhancement of Transient Stability for
Iraqi Power system by Using Fuzzy Genetic Controller, Ph.D Thesis,
university of Technology, Baghdad, 2007.

311